AMD once again violating power specifications? (AMD RX-480)

[Godlygamer23]

18 minutes ago, Robin88 said:

If I understand this correctly, and you can absolutely correct me if I'm wrong.

But according to Toms Hardware, the entire card is pulling at peak, some 300W of power? And at that peak power draw, some 155W to 200W of that is coming from the PCIe connector on the motherboard? If that is true, then holy cow AMD, how on earth did that pass validation?

But here's the thing, I call bullshit on Toms numbers and here's why. The freaking motherboard would have fried with that much power being pulled through something only meant for 75W MAX. There is no motherboard on earth without additional power inputs that could cope with that kind of power draw, none.

Why do you think that some mobos have an extra molex connector next to the PCIe slots? Because PCIe by itself cannot cope with more than a peak power draw of 100W per slot, and definitely no more than 75W long term.

They might be able to if it's only peak for a second or less. I don't remember if the review mentioned the amount of time it stayed at that value for the 155W number. I agree it's an insane number, but peak does not mean continuous. To me though, violation of a spec is enough, especially since this is a mainstream card. I don't care much about enthusiast cards like the 295X2, but if it's a budget card, and meant for budget gamers, that also means the motherboard and power supply will likely not be super amazing. 

[Misanthrope]

5 minutes ago, mariushm said:

@Robin88 , those peaks are for VERY SHORT periods of times, something like 1 ms or maybe a few ms.  The average over the course of let's say 200 ms is about 80 watts.

 

See for example the GTX1080 review, where the card pulls over 250 watts from the pci-e 8pin connector alone for short periods of time (it averages at less than 150 watts) : http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1080-pascal,4572-10.html

 

 

 

 

 

Yet at the same time, if lighting strikes you down it will similarly be only a few milliseconds of your body getting the very high current, yet you still get roasted to death.

[Robin88]

11 minutes ago, mariushm said:

@Robin88 , those peaks are for VERY SHORT periods of times, something like 1 ms or maybe a few ms.  The average over the course of let's say 200 ms is about 80 watts.

 

See for example the GTX1080 review, where the card pulls over 250 watts from the pci-e 8pin connector alone for short periods of time (it averages at less than 150 watts) : http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1080-pascal,4572-10.html

 

 

 

 

 

Thanks for that, I stand corrected, fuck AMD, once these GPU's get into peoples hands, I wouldn't be surprised if we see mass mobo and PSU deaths.

I'll wait and see what Anadtech says, but as it stands, I won't be recommending the RX480 until there is either a recall, or AMD releases some kind of software/firmware fix that can correct that kind of power draw from the slot.

Geez AMD, the first GPU release since the HD5000 series that has had me properly excited and you go and fuck it up. Well done.

nVidia don't deserve a free pass either btw, but at least with it happening on the 1080, it's generally bought by people who can afford quality parts, whereas the RX480 is more likely to be bought by those who are on a budget and are likely to skimp on the mobo to save money, and are more likely to suffer dead boards due to this.

[DevilishBooster]

5 minutes ago, Godlygamer23 said:

They might be able to if it's only peak for a second or less. I don't remember if the review mentioned the amount of time it stayed at that value for the 155W number. I agree it's an insane number, but peak does not mean continuous. To me though, violation of a spec is enough, especially since this is a mainstream card. I don't care much about enthusiast cards like the 295X2, but if it's a budget card, and meant for budget gamers, that also means the motherboard and power supply will likely not be super amazing. 

Just for my clarification, are you considering it a violation of spec if the power exceeds it for even a fraction of a second? Cause if so, then Nvidia is guilty of it as well because as the graphs posted previously show, the 1080 also has momentary spikes in power draw that violate the standard.

[Misanthrope]

2 minutes ago, DevilishBooster said:

Just for my clarification, are you considering it a violation of spec if the power exceeds it for even a fraction of a second? Cause if so, then Nvidia is guilty of it as well because as the graphs posted previously show, the 1080 also has momentary spikes in power draw that violate the standard.

The standard probably indicates how long and how often would be considered no longer compliant or if it's an absolute limit. Can't look this up at this time though.

[Trixanity]

6 minutes ago, Misanthrope said:

Yet at the same time, if lighting strikes you down it will similarly be only a few milliseconds of your body getting the very high current, yet you still get roasted to death.

TIL I'm a motherboard.

[ace_cheaply]

Well, @Majestic, in the face of overwhelming evidence, it comes to me that you might just be right about this.  Maybe.

 

I still probably will  get a Nitro or a Toxic if they make it, and it overclocks to 1400.  I have a Asus Sabertooth z170s, so I should be fine.  my 6700k will make sure I get the most out of the 480, and I just don't have the budget for a 1070.  I either have to go used, a 480, or the 390x nitro for $300 which is stretching my budget more than I'd like.  Although my 770 doesn't do bad enough that it has to be immediately replaced.  I could wait until the 490, i am just itching. 

10 minutes ago, DevilishBooster said:

Just for my clarification, are you considering it a violation of spec if the power exceeds it for even a fraction of a second? Cause if so, then Nvidia is guilty of it as well because as the graphs posted previously show, the 1080 also has momentary spikes in power draw that violate the standard.

It has been pointed out that the 1080 draws it through the connector, and not through the pci-e slot. 

[mariushm]

2 minutes ago, DevilishBooster said:

Just for my clarification, are you considering it a violation of spec if the power exceeds it for even a fraction of a second? Cause if so, then Nvidia is guilty of it as well because as the graphs posted previously show, the 1080 also has momentary spikes in power draw that violate the standard.

Tom's Hardware argues that the fluctuations in current going through the motherboard to the pci-e slot could be "mirrored" or "reflected", something like that into audio paths, potentially causing noise on very cheap motherboards when scrolling (because video card would suddenly pull more current as whole screen changes suddenly) or when gaming. But we're talking stupid cheap or badly designed motherboards, boards that would normally be paired with under $100 video cards.

It's not about the tiny difference of a few watts between the maximum allowed by the pci-e standard and what the card draws, it's about the tiny ms long bursts when card can pull much more than 60w from the slot.

 

[Prysin]

14 minutes ago, Majestic said:

Defining what is classified as high quality. It's not as simple as "the most expensive", or "brand A or B".

Very well said.

Great example was first batch of FuryXs. Intrusive coil whine.

 

Then there was the GTX 580. High end for its time but rampant thermal throttling and overheating.

 

Gigabyte R9 300 series cards came with broken fans or missing fan profiles.

 

Asus R9 300 series cards had large number of broken vBIOS on arrival.

 

GTX 970. Needless to say more.

 

Paying more is never a guarantee 

[Prysin]

5 minutes ago, ace_cheaply said:

Well, @Majestic, in the face of overwhelming evidence, it comes to me that you might just be right about this.  Maybe.

 

I still probably will  get a Nitro or a Toxic if they make it, and it overclocks to 1400 though.  I have a Asus Sabertooth z170s, so I should be fine.  my 6700k will make sure I get the most out of the 480, and I just don't have the budget for a 1070.  I either have to go used, a 480, or the 390x nitro for $300 which is stretching my budget more than I'd like.  Although my 770 doesn't do bad enough that it has to be immediately replaced.  I could wait until the 490, i am just itching. 

It has been pointed out that the 1080 draws it through the connector, and not through the pci-e slot. 

But the 1080 is Nvidia. So naturally it's not a problem when they break the "standards"

[Kobz360]

Man this is tragic and isn't too good to hear at all. I'd like to hope that AMD can somehow fix this issue (I'm not very knowledgeable in this area), otherwise it could be problematic. I have a question though, when AIBs utilise the 8-pin, does that mean this issue will no longer be as significant or? 

 

Regardless, I really hope nobody's board gets fried but we'll see if anything happens to anyone as time progresses 

[NumLock21]

23 minutes ago, Majestic said:

Defining what is classified as high quality. It's not as simple as "the most expensive", or "brand A or B".

True, you have server boards that cost way less, i doubt they're not high quality.

[raphidy]

9 minutes ago, DevilishBooster said:

Just for my clarification, are you considering it a violation of spec if the power exceeds it for even a fraction of a second? Cause if so, then Nvidia is guilty of it as well because as the graphs posted previously show, the 1080 also has momentary spikes in power draw that violate the standard.

I don't know standards for electronics, but usually, it's about continuous loads not peaks. Average load is continuous load which is far more dangerous than peaks. Of course, you also need your component resist some sort of peaks. I would better have the psu handle the peaks than the motherboard though.

[ace_cheaply]

9 minutes ago, Prysin said:

But the 1080 is Nvidia. So naturally it's not a problem when they break the "standards"

Naturally not, since they are flawless and perfect and wonderful.

That said, I'm not overly attached to any brand.  I'll admit that AMD holds a special place in my heart for the 4870, the first video card that I bought myself, which still runs in my htpc. I did have a gtx 8800 before that, but it was given to me and it didn't have the same sentimental value.  However, I've gone back in forth for generations. 4870, 560ti, hd 6950, HD 7870, GTX 770, I just go with whatever I can get the price to performance at the time.  I plan on getting a 480, and finally getting a tier 1 card again when vega/big pascal hits.  

 

[Dabombinable]

18 minutes ago, Prysin said:

But the 1080 is Nvidia. So naturally it's not a problem when they break the "standards"

They are breaking the ATX standard the way AMD did with the 295 X2-through the PCIe power connectors. Not the motherboard which is the issue at hand.

[spartaman64]

16 hours ago, Majestic said:

Don't expect a plethora of reviewers running the same kit, as was stated earlier. If a test is done properly, with care, I don't see how it's invalidated simply for being the only one reporting it.

they might have gotten a defective card. their testing equipment might have been defective. there are a lot of factors that could have produced incorrect results so we should wait for more tests before actually confirming this

[Majestic]

35 minutes ago, Prysin said:

But the 1080 is Nvidia. So naturally it's not a problem when they break the "standards"

Why Prysin, why the sneer. Anyway, 

 

 

Average 150W, it's a 8pin. No violation. It's also during OC. No OC it's about 130-ish

[Majestic]

3 minutes ago, spartaman64 said:

they might have gotten a defective card. their testing equipment might have been defective. there are a lot of factors that could have produced incorrect results so we should wait for more tests before actually confirming this

You are in luck, check the OP again. Lawlz added a few.

[FatalityDiablo]

So an 8pin aib model would solve the problem?

[Senzelian]

Just another reason to stick with NVidia... ?

[spartaman64]

7 minutes ago, Majestic said:

You are in luck, check the OP again. Lawlz added a few.

also the rx 480 had to pass the PCI-SIG's own testing to be released to market and apparently it did so both amd and PCI-SIG messed up their testing?

[Dabombinable]

2 minutes ago, spartaman64 said:

also the rx 480 had to pass the PCI-SIG's own testing to be released to market and apparently it did so both amd and PCI-SIG messed up their testing?

It was probably at a lower clock speed to pass.

[spartaman64]

5 minutes ago, Dabombinable said:

It was probably at a lower clock speed to pass.

i dont think they would have allowed that and amd is apparently trying to recreate reviewer's issues internally with their engineering team

[KeltonDSMer]

45 minutes ago, Godlygamer23 said:

They might be able to if it's only peak for a second or less. I don't remember if the review mentioned the amount of time it stayed at that value for the 155W number. I agree it's an insane number, but peak does not mean continuous. To me though, violation of a spec is enough, especially since this is a mainstream card. I don't care much about enthusiast cards like the 295X2, but if it's a budget card, and meant for budget gamers, that also means the motherboard and power supply will likely not be super amazing. 

Can you tell me how the 12V traces from 24pin to PCIe socket are wired in a mobo? I was under the impression that they are direct from the 24pin to the PCIe socket...no components to burn out except the traces themselves. If this is the case, why is this an issue? Drawing power via the slot would be no risk to the board unless the PCB itself can't dissipate a watt or two from the voltage drop over the length of the traces. The risk is further lessened from the fact that these are peaks and not RMS...not very much heat to get rid of if the traces are in fact direct.

 

Is this not true, does the mobo "condition" 12V in some way?

[zMeul]

13 minutes ago, FatalityDiablo said:

So an 8pin aib model would solve the problem?

no, why?

• the difference between a 6 pin PCIe connector the the 8 pins on is added 2 ground wires

• the problem with RX480 is that it draws more power from the PEG slot, meaning there is no Over Current limiting circuitry, or if it exists it doesn't do it's job

 

 

it could be fixable by a custom AIB partner is they design the power delivery circuitry with PEG power limit in the 1st place

Edited June 30, 2016 by zMeul